Methodological Limitations

• Funding conflict of interest: The work was commissioned by the Society for Evidence-based Gender Medicine (SEGM), an organization with a known critical stance on gender-affirming care for youth, creating potential sponsor bias in study selection, framing, or interpretation.
• Direct financial compensation to authors from SEGM: RB-P and AM 'provided methodological expertise for the SEGM initiative to summarise and appraise the quality of publications related to gender medicine for the SEGM online platform, and for this work they received financial compensation from SEGM.'
• Protocol authors with SEGM affiliations: William Malone is a board member of SEGM and a contributing author for the SEGM online platform, and participated in generating the systematic review question, potentially biasing the research question formulation.
• Language restriction bias: The review included only English-language studies, which may exclude relevant non-English research and systematically bias findings.
• Modified risk of bias tool without validation: The authors used a 'modified version of the Cochrane risk of bias tool for non-randomised studies of interventions (ROBINS-I)' without demonstrating that this modification maintains validity or reliability; they merely asserted it 'produced similar conclusions.'
• Inconsistent handling of heterogeneity: The authors chose not to rate down for inconsistency despite considerable heterogeneity (I²=88%, p<0.01 for global function; I²=94%, p<0.01 for global function in before-after studies; I²=100%, p<0.01 for depression) based on post-hoc explanations about sex-specific measurement rather than prespecified criteria.
• Implausible precision in BMD estimates: Hip BMD was reported as '0.01 higher (95% CI 0.01 higher to 0.01 higher)' with identical point estimate and confidence limits, suggesting computational or reporting error rather than genuine precision.
• Contradictory certainty ratings for cardiovascular events: The same outcome (cardiovascular events in NFs) received 'high certainty' in one case series and 'moderate certainty' in another, with no clear justification for this discrepancy beyond differing follow-up periods and event definitions.
• Circular reasoning in case series certainty assessment: The authors rated cardiovascular events as 'high certainty' in case series because 'this outcome does not need a comparison group,' yet cardiovascular events in intervention recipients inherently require comparison to establish causality or even baseline risk.
• Post-hoc reclassification of study design: One before-after study was 'classified as a case series' for the suicide outcome because it only reported post-intervention data, suggesting selective reclassification based on data availability rather than prespecified criteria.
• Failure to define minimally important difference: The authors 'did not define a minimally important difference to determine whether an effect was clinically meaningful or important,' potentially obscuring clinically relevant effects.
• Imputed correlation coefficient without sensitivity analysis: The authors 'imputed a moderate correlation coefficient (r=0.5)' for before-after studies without reporting sensitivity analyses to assess robustness to this assumption.
• Selective outcome prioritization: Due to 'feasibility considerations,' the authors excluded potentially important outcomes including 'regret, anxiety, pelvic pain or cancers,' which may be highly relevant to harm assessment.
• Potential confounding by indication: The comparative observational studies showed serious risk of bias from confounding, with lack of adjustment for 'psychiatric interventions, mental health comorbidities, socioeconomic status and family support'—factors strongly associated with both treatment access and outcomes.
• Co-intervention contamination: In one study, '28.26% of the participants in the no GAHT group were receiving puberty blockers or spironolactone as monotherapy,' severely compromising the comparison.
• Extreme missing data: Multiple studies had critical risk of bias due to missing data, including 41.71%, 46.75%, 48%, 69% of participants not providing outcome data, threatening validity of conclusions.
• Inadequate sample sizes: The authors noted that 'optimal information size (200) was not met' for some meta-analyses, yet proceeded with pooled estimates despite high risk of random error.